AUTOMATIC TEST GENERATION BASED ON CONSTRAINTS

Abstract

It seems to be a very hard task to enhance the properties of widespreadly used automatic test pattern generation algorithms. Experiences show that achievements are sometimes not worth the effort. In the authors' opinion this fact stems from the basically 'algorithm oriented' nature of research made in the past. A new experimental framework is presented for the problem, considering network representation and search control algorithms as equally important parts. The network is represented by object- oriented data-flow networks, the search control algorithm is based on constraint satisfaction, and a special kind of dependency directed backtracking which we call constraint slackening. Similar methods were proved to be very useful in automatic system diagnosis by DAVIS (1985) and others, although have not been introduced to testing yet. This paper summarises the basic notions of constraint satisfaction, the potential advantages of using it for building test generation systems, and shows implementational details of a test generation system, based on constraints. Experiences of the run-time tests show that constraint-based test generation can be highly efficient.